Mass Wasting Identification in the Navarro River Watershed Using Hyperspectral Imagery

Abstract

Several of northern California's coastal watersheds have been listed by the U.S. Environmental Protection Agency as requiring the development of a Total Maximum Daily Load (TMDL) for non-point source pollution. One such case is the Navarro River watershed located in southern Mendocino County. The Navarro River was listed under the federal Clean Water Act Section 303(d) for the constituents of temperature and sediment. The Information Center for the Environment (ICE) at the University of California, Davis aided the North Coast Regional Water Quality Control Board in the development of the TMDLs and continues to assist with ongoing work in the Navarro River watershed and other North Coast watersheds. During the summer of 2000, hyperspectral imagery was collected for the entire Navarro River watershed (∼820 km 2 ) using the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) administered by the National Aeronautic and Space Administration. This imagery covers the spectral wavelength range from 400 to 2500 nm with spectral sampling of 10 nm, 224 contiguous channels, and a spatial resolution of 5 meters. Potential mass wasting sites were extracted from the AVIRIS imagery using various image processing techniques, such as the Tasseled Cap transformation, image segmentation and masking. Spectral unmixing algorithms were employed for mixed pixels (vegetation and soil) to determine the percentage of each material per pixel. Augmenting these data within a Geographic Information System (GIS), a digital elevation model was used to determine proximity to stream systems and hillslope position. This study focused on features in one sub-watershed, predominately on private timber lands: the North Fork of the Navarro River. The AVIRIS data analyzed for this area is of particular interest because it is one of the few sub-watersheds of the Navarro River complex that provides suitable spawning habitat for threatened coho salmon ( Oncorhynchus kisutch ). Additionally, substantial fieldwork was performed in this sub-watershed to verify the spatial position and dimensions of identified mass-wasting sites. Of these approximately 1100 manually mapped and digitized mass-wasting sites, one-half were used for assessing commission and omission errors of features extracted from the AVIRIS imagery. The other sites were used to assess the accuracy of the terrain-based model SHALSTAB, which predicts landslide potential. Hyperspectral data do provide a means for the inventorying and monitoring of sediment delivery sites over large areas; however, the geometric and atmospheric corrections required to effectively process these data can be onerous. The utilization of hyperspectral imagery for watershed analysis is novel and holds great promise for future TMDL assessments, but it is not without limitations. The sheer volume of data, over 300 Gigabytes for the Navarro River watershed alone, can be overwhelming to an ill-equipped processing laboratory. Furthermore, comprehensive fieldwork is necessary for more advanced applications of spectral separability and identification. As hyperspectral data are incorporated into future watershed research programs, many of these benefits and limitations will be addressed. Although the identification of mass-wasting features was performed post-hoc for the Navarro River watershed, our research indicates that similar approaches can be used for other watersheds for a variety of purposes with success.